llama.cpp b9830 release adds --offline flag and fixes memory bug

The llama.cpp b9830 release introduces the ability to use the --offline flag with the llama download command, allowing scripts to verify cached models without network access. This update also resolves a latent use-after-free vulnerability in the URL-task on_done callback where first_path was incorrectly captured by reference.

• Exposes the existing --offline flag to llama download for offline model caching checks.
• Fixes a use-after-free bug in the URL-task on_done callback involving block-scoped variables.
• Provides binaries for macOS (Apple Silicon and Intel), iOS, Linux (CPU, Vulkan, ROCm, OpenVINO, SYCL), Android, Windows (CPU, CUDA 12/13, Vulkan, OpenCL, OpenVINO, SYCL, HIP), and openEuler.

This release enables safer offline workflows for model management while addressing a critical memory safety issue in the download functionality.